I’m looking to train a Face Detector with Facial Landmarks, but I’m having issues using the jetson-inference examples to learn from.
I’m running on a Jetson Nano.
I have my dataset of images and annotations, but I noticed that the models that are specified in the
train_ssd.py script in
jetson-inference are tailored to pure object detection.
DetectNet has the option to run a
facenet-120 model, which runs great, but is there a way to train this model myself and somehow include Facial Landmarks?
My goal is eventually to be able to run detectNet with my freshly trained network that can detect not only faces, but landmarks that I have specified.
Alternatively, I could probably achieve this with a Keypoint Detector. Is there a way to train a keypoint detector using the
train-ssd script? Or do all of the available
--net options only allow for a dataset of bounding boxes rather than keypoints?